When the first version of Sora was released in February 2024, it felt like the GPT-1 moment for AI video. Anyone can generate highly realistic videos with nothing but text prompts. However, the results weren’t quite production-ready. Subjects often looked physically off, and there was no native support for audio or sound effects.
Sora 2 is a GPT-4 moment for AI video. The upgrade is massive.
It solves some insanely difficult motions like Olympic gymnastics routines, backflips on a paddleboard that accurately model the dynamics of buoyancy and rigidity, and triple axels while a cat holds on for dear life.
I’ll go straight ahead and show you one good example of how huge the improvements are.
Prompt: A man does a backflip on a paddleboard